Connecting Grammatical Person, Humanness, and Discourse Functions: Always Progressives as a Case Study

Abstract

Because they reflect such important pragmatic and cognitive aspects of discourse, grammatical person and humanness merit special consideration in the analysis of any linguistic structure containing nominals. Progressive verbs tend to take human and volitional subjects, a pattern one might predict extends to always progressives used as complaints, e.g., you’re always muttering. (Being agentive, humans can be held accountable for irritating actions.) This functional, corpus-based study explores the relationship between the functions of 1233 always progressives and features of their grammatical subjects. Tokens were coded by subject type (first person pronoun [1PP], addressee you [Addr], generic you [Gen], third party [3P], or nonhuman [NH]) and function (complaint, lament, positive evaluation, or emotionally neutral description). A configural frequency analysis indicated that the combinations 3P-Complain, 3P-Positive, and Addr-Complain were more frequent than expected and the combinations 1PP-Complain, NH-Complain, Gen-Complain, 1PP-Positive, and 3P-Describe were less frequent. I propose sociological explanations for these findings, such as a reluctance to brag and the fact that gossip—the sharing of negative information about third parties—serves crucial social functions. In addition, I show that generic you is semantically and functionally similar to the 1PP, and that, given this, analyses should not conflate generic and addressee you.

Introduction

Grammatical person, encoded explicitly in pronouns, reflects an important socio-pragmatic aspect of the discourse context. Utterances about ourselves tend to serve different purposes than utterances about others. For example, we typically would not make out-loud requests of ourselves, as we already know our own wishes, and saying good things about oneself (bragging) is a very different act than saying good things about someone else (praising/complimenting). The same is true of utterances about humans as opposed to inanimate objects. The significance of these basic categories of nominals (e.g., human versus nonhuman and self versus other) is recognized in Functional Linguistics, where much careful work has been done on humanness, grammatical person, animacy, and more. This includes work relating such features to syntax, primarily to word order. However, empirical studies that relate the functions of specific constructions to the types of nominals they contain are rare. The present work demonstrates the explanatory power of this underutilized variable by incorporating it into an empirical analysis of the nominals that serve as the grammatical subjects of always progressives.

The progressive is the focus of a large body of research in English linguistics aimed at determining how it evolved and establishing taxonomies of its uses and meanings, but there are few corpus analyses of its modern usage, and the way we use progressives containing the adverb always (e.g., you’re always muttering) has been misunderstood. Numerous descriptive grammars state that always progressives tend to be complaints or otherwise negative, offering examples such as He’s always moaning about money (Carter & McCarthy, 2006: 47). Always progressives certainly can sound like complaints, particularly if the subject is human and second or third person. However, Lindley (2020) analyzed 752 such progressives and found that only 16.9% were complaints. The majority (71.9%) served as neutral descriptions, and the rest were laments or praise.

Based on a new and expanded data set, the present functionalist investigation of always progressives relates their functions to one of their key components: the grammatical subject. The relationship between the function and grammatical subject type of 1,233 always progressives was tested using a configural frequency analysis, which identified the subject-function combinations with the strongest negative or positive relationships. The paper is structured as follows: The next section discusses relevant literature. I then describe the method and results. In the discussion section, I offer explanations for the discovered usage patterns, and the paper ends with concluding remarks.

Background

The existing research relevant to this study is that which focuses on agentivity, person, humanness, and other tightly interrelated features of nominals. This body of work spans three major strands of functionalist (non-generative) linguistics—Functional Linguistics, Cognitive Linguistics, and corpus linguistics—and, in accordance with the interests of the researchers, treats those features as pragmatic (discourse-situational), semantic, or cognitive. Also relevant are corpus studies that determine which items appear in constructions and corpus studies of the progressive. Few studies combine these research areas, though, by relating the functions of a construction to the nominals it contains. This section begins by reviewing the aforementioned literature and then discusses the progressive and always progressives.

Person, Humanness, and Related Features

In Functional Linguistics, nominals have long been written about in terms of a cluster of semantic features related to agentivity: humanness, animacy, and volitionality. For example, Hopper and Thompson’s (1980) transitivity hierarchy involves agentivity (or agency) and volitionality. The former is a characteristic of nominals and the latter a characteristic of a verb/action, but the two are tightly connected. Hopper and Thompson equate volitionality with “acting purposefully” (252) and “voluntary participation” (256)—referring, notably, to both an action and the mindset of the doer of that action.¹ The notions are separable in that agentive subjects can appear with non-volitional verbs (such as forget), but overlap in that “a volitional verb requires an agentive subject—one that is human, or at least animate” (286). This same logic is reflected in Comrie’s (1989) animacy hierarchy consisting of the ranking human > animal > inanimate² and Corbett’s (2000: 56) very similar animacy hierarchy. The idea is that humans are able to engage in actions purposefully while inanimate things such as rocks are not, and animals (animate but not human) fall in between: Their capacity to consciously choose to do things is typically viewed as a weaker version of humans’. Similar orderings are posited by Silverstein (1976) regarding ergativity and by Kuno and Kaburaki (1977) regarding empathy. In sum, much research in Functional Linguistics indicates that language use is affected by features of nominals such as humanness and animacy and/or by the cognitive capacities (the ability to act with agency/commit volitional acts) tied to such features.

Nominals are also analyzed in Functional Linguistics in terms of a pragmatic feature called topicality (Givón, 1976). Nominals that serve as topics have the status of old/given information that the rest of the sentence tells us more about. For example, in the exchange “How’s Olga?”–“She’s doing well,” the pronoun she refers to a person who has been established as a topic, and we learn the additional information that she is doing well. This status affects syntactic structure in that topics tend to appear earlier in utterances and to be grammatical subjects. Topicality is entwined with agentivity in that agentive things—because of their ability to (potentially purposely) engage in actions that affect our lives—presumably are likely to be the topic of talk. This explains why Comrie’s “animacy” consists of not just literal animacy but also “topic-worthiness” (1989: 199), why humanness is found in Givón’s topicality hierarchy (1976: 152), and why both topicalization and animacy figure in Timberlake’s participation individuation hierarchy (1975: 134).

Topicality is simultaneously a cognitive feature, equivalent to what in Cognitive Linguistics is called cognitive salience (Taylor, 1994: 72; Langacker, 1991: 132). Concepts are cognitively salient if they have been “activat[ed] … in actual speech events” and are thus part of one’s current focus (Schmid, 2007: 119). A similar cognitive feature of nominals is ontological salience, defined as the inherent “attention-attracting potential” of the referent of a nominal (Schmid, 2007: 120). This, too, overlaps with agentivity and topicality. Agentive things are inherently attention-grabbing³ (i.e., ontologically salient), and attention-grabbing things often truly become someone’s focus of attention (i.e., become topics, become cognitively salient). The same is true of things that are individuated as opposed to being perceived as part of a mass or group. The interrelatedness of topicality, salience, and agentivity explains why Timberlake relates individuation to topicalization, humanness, and animacy and why individuation appears in Givón’s topicality hierarchy (1976: 152), Comrie’s definition of animacy (1989: 199), and Hopper and Thompson’s transitivity hierarchy (1980). Thus, salience of one type or the other captures much of what has been studied in Functional Linguistics.

The final relevant well-studied feature of nominals, one that is both pragmatic and cognitive, is what Kuno (1987) calls “empathy.” Similar to point of view or “a camera angle,” empathy is “the speaker’s identification, which may vary in degree, with the person/thing … in a sentence” (206). It equates to “subjectivity” in Cognitive Grammar: The closer an entity is to being the conceiver rather than the thing conceived, the more subjectively it is construed, meaning that “the crucial factor is vantage point” (Langacker, 1999: 149) or “perspective” (150). In pronouns, this is encoded explicitly via grammatical person: First person pronouns (1PPs) signal a point of view identical to the speaker’s/conceiver’s. Thus, in terms of empathy, 1PPs clearly outrank all other nominals (Kuno, 1987: 212). In contrast, speakers identify the least with 3PPs because they refer to entities that, unlike the speaker and 2P addressee(s), are not even participating in the interaction.

Empathy and grammatical person are closely linked to all the features discussed thus far. Kuno writes that topics are empathized with as easily as or more easily than nontopics (1987: 210), and Langacker considers empathy to be a component of topicality (1991: 306). Kuno and Kaburaki’s humanness hierarchy, regarding empathy, consists of human > animate nonhuman > thing (1977: 654), and person is prominent in both Givón’s topicality hierarchy (1P > 2P > 3P) (1976: 152) and Corbett’s Animacy Hierarchy (1PP > 2PP > third person > kin > human > animate > inanimate) (2000: 56). These rankings seem to treat empathy in both the camera angle sense, which explains the pronoun rankings, and in a more everyday sense, which explains the presence of terms such as human, kin, animate non-human, and thing. (We more easily identify with fellow humans than with animals and identify even less with things, such as chairs.) The present study continues this functionalist tradition of studying such features—but does so with the goal of understanding a specific construction and through using corpus methods.

When nominals are analyzed in corpus linguistics, it usually is in terms of semantic prosody and/or semantic preference. The semantic prosody of a word is the “consistent aura of meaning”—restricted to the semantic field of pleasantness: negative (unpleasant), positive (pleasant), or neutral—“imbued by its collocates” (Louw, 1993: 157). Collocates are words that frequently appear near the word of interest. For example, break out (intransitive) has negative semantic prosody because it appears with grammatical subjects such as disorder and epidemics (Sinclair, 1990: xi). A similar concept is a “collexeme” (Stefanowitsch & Gries, 2003: 215), a word that frequently appears in a particular construction. A collexeme along with the grammatical construction associated with it is called a collostruct (Stefanowitsch & Gries, 2003: 215). This construction-focused approach to corpus analyses is compatible with a large set of approaches to linguistics that began to emerge around the late 1980s and that blur the line between lexicon and grammatical structures. The authors associate it in particular with the Construction Grammar of Lakoff (1987) and Goldberg (1995). The central idea of such approaches is that not just morphemes but grammatical constructions can be “directly associated with a particular meaning or function” (Stefanowitsch & Gries, 2003: 212). Constructions can be as small as a morpheme or as large as a multi-word expression (in which perhaps not all the words are specified), and their “form or meaning cannot be compositionally derived” (ibid.).

In corpus linguistics, words (whether treated as collocates or collexemes) can also be profiled in terms of semantic preference. This is similar to semantic prosody but flexible regarding the semantic field and therefore able to include features like agentivity. For example, sleep attracts grammatical subjects that are animate. Work of this nature has been conducted on the grammatical subjects of the most common progressive verbs and has revealed that they tend to be human (Biber et al., 1999: 473).

Always Progressives

The Present Day English progressive has two main types of meanings, aspectual and subjective. Its aspectual meanings are, roughly, that the action/state is in progress, continuous, and/or repeated, while its subjective meanings are those “based in the speaker’s subjective belief/attitude toward the situation” (Traugott, 1990: 500)⁴—such as being irritated by it. The aspectual meaning is more common,⁵ but originally the construction had only subjective meaning. Specifically, it was used mainly for emphasis (Kranich, 2010: 163).

Analyses of the progressive (Killie, 2004; Wright, 1994; and Smitterberg, 2005) tend to present ALWAYS⁶ progressives as the prototypical example of the subjective progressive. Moreover, various sources state that ALWAYS progressives typically convey a negative attitude or emotion (e.g., irritation or disapproval) and/or are used for critical evaluations (e.g., Carter & McCarthy, 2006; Edmondson & House, 1981; Jespersen, 1961 [1931]; Kranich, 2010; Leech, 2004; Sinclair, 1990; Smitterberg, 2005; Quirk et al., 1985). In other words, they are usually complaints. An example from Killie (2004: 40) is he’s always saying ‘I’ll see about it’—and he never does see. Some concede that ALWAYS progressives have non-negative uses, such as conveying emotion in general (Smitterberg, 2005) or even conveying approval (Celce-Murcia & Larsen-Freeman, 1999), as in He’s always delivering in a clutch situation (ibid.: 117; cf. Kranich, 2010: 66). Finally, Römer (2005) and Kranich (2010) show that always progressives need not even be subjective. More commonly, though, their non-negative uses are minimized or not even mentioned.

ALWAYS progressives do seem conducive to complaints. Words like always are associated with hyperbole (Claridge, 2011: 51), which is associated with emotionality and negative evaluations (162). Meanwhile, the progressive -ing causes us to construe the event/state as homogenous and, when combined with auxiliary be, “profiles an internal portion” of a process that “excludes the endpoints” (Langacker, 2008: 155). It is as if one is placed in the middle of something monotonous with no end in sight, and if always is included, there is not even the hope of an end—making for an unpleasant experience.

Nevertheless, there is no evidence that the complaint function of ALWAYS progressives is the most common. Kranich (2007: 130, 2010: 65) states that ALWAYS progressives tend to be interpreted negatively, yet when reinterpreted in terms of positive, negative, or neutral emotions (see Lindley, 2020: 338, Table 1), her two corpus studies indicate a strong link between ALWAYS progressives and emotion but a weaker link between them and negative emotion. Notably, Römer’s (2005) monograph on progressives discusses a hitherto overlooked meaning of the progressive that is emotionally neutral and likely linked to always: the “general validity” meaning, used “when something is true or valid in general, or at least for some time” (96). About 22.5% of progressives in her study had this meaning (95), and the most common time adverbials in the progressives were when, always, and all the time (97).

Römer’s general validity progressives help explain the findings presented in Lindley (2020). I analyzed 752 Present Day English always progressives using a coding schema consisting of four functions: Complain, Lament, Praise, and Describe. In the study, the emotionally neutral function (Describe) accounted for 71.9% of the tokens, whereas only 16.8% were complaints. I argue that the linguistic features that make the construction useful for complaints are equally useful for the other three functions. For example, the semantics of permanent, ongoing action lend themselves well to neutral statements of facts. Such always progressives are synonymous with Römer’s general validity progressives.

The distribution of functions across always progressives (of all types) though, is only one piece of information about them. A linguistic component of the always progressive construction that is yet to be explored in relation to function is its grammatical subject. It is known that the most commonly used progressive verbs “typically take a human subject as agent […], actively controlling the action (or state) expressed by the verb” (Biber et al., 1999: 473). This link appears to have been even stronger in the past: Kranich writes that until the 18th century, “the progressive occur[ed] predominantly with agentive (mostly [+human]) subjects” (2010: 118). This tendency may or may not extend to always progressives.⁷

Supposing that it does, we can note that such subjects, being agentive and cognitively advanced, are more likely than others to have actions attributed to them and can more plausibly be held responsible for those actions—yet the construction is not used as a complaint especially often. However, this does not indicate that the features of the grammatical subject are irrelevant to the complaint function; it may just be that we need to differentiate the types of human subjects. A feature that could be particularly useful to explore and which figured prominently in the Functional Linguistics work described above is grammatical person. For example, we probably complain about others more than ourselves, so 1PP subjects may be rare in always progressives serving as complaints.

The present study tests if such subject-function associations exist. That is, it relates the semantic preferences of the nominal collexemes of always progressives to the functions of that construction. The value of such a study is hinted at in Römer’s comment that there are connections between general validity progressives (a construction with a certain function), their grammatical subjects, and their adverbials (2005: 96). Specifically, I test the relationship between a subject’s status as human and first, second, or third person or as nonhuman and the function of the always progressive containing it. Doing so unites the functionalist interest in features related to agentivity, empathy, and so on with corpus techniques for analyzing collexemes. While we have some information about the typical subjects of progressives, we know less about those of always progressives, and no study has linked their subject types to their functions.

Method

Always progressives were collected and coded according to (a) their discourse functions (Describe, Lament, Complain, Evaluate Positively) and (b) features of their grammatical subjects (person and humanness). The tokens were in past or present tense and with the auxiliary verb in pre-adverb position, matching the relevant examples in the literature. To limit the effect of confounding variables, I considered only always rather than always and its synonyms.

Data

This study makes use of the Corpus of Contemporary American English (COCA) (Davies, 2008–), a balanced monitor corpus of modern American English from 1990 through 2019. COCA at the time drew equally from five genres: speech, newspapers, magazines, fiction, and academic journals. The analyzed tokens consisted of every always progressive in the 2014–2017 data, which contains 82,236,871 words.

Token Collection

Using the List option of the corpus’s online interface, for each year, I searched for “always [v?g*],” where “[v?g*]” denotes the -ing form of a verb.⁸ The linguistic context (KWIC view) was collected for every hit as well as the expanded linguistic context (about 160–180 words).

I manually rejected hits that did not meet the criteria of being an always progressive in the (a) declarative mood,⁹ (b) simple aspect, (c) active voice, and (d) past or present tense; (e) not negated; with an auxiliary verb that is (f) overtly present and (g) in pre-adverb position; and (h) with an overtly present subject. Tokens also were required to have nothing intervening between (i) the subject (including its modifiers, if present) and auxiliary verb or (j) the auxiliary verb and always. (Real examples of such intervening material are for instance, as you say and modifiers of always such as almost and perhaps.) In addition, three criteria (k-m) involving word order were used: (k) direct objects cannot precede the verb and must be overt, (l) objects of prepositions cannot precede the verb (unless the entire prepositional phrase does so) and must be overt, and (m) the verb cannot precede the subject (something that occurred twice).

Hits were also rejected for the following reasons: (n) They were duplicates. By this I mean instances—whether or not flagged by COCA as such—in which an always progressive appears multiple times for a reason other than it truly being produced multiple times, such as when a television show plays a video clip twice. (o) They were erroneous transcriptions of audio, as verified through watching the original video. There was only one instance of this. (p) They were not originally from 2014 to 2017, as was the case regarding a token that was a quote from a play written in 1607. In over a dozen instances, the search results erroneously included tokens from years other than the year selected. If such tokens were nevertheless from 2014 to 2017 and met the above criteria, I retained them. While checking each hit manually, I collected any always progressives that appeared in the extended context yet had failed to appear as hits themselves. This happened twice.

Failure to meet criterion (f) was the sole reason for nearly half of all rejections, the vast majority of which involved participial phrases or gerunds. Criterion (d) accounted for the second most rejections (about a quarter), nearly all of which involved the go-future.

Coding of Function

Each always progressive was coded manually according to its discourse function. Examples from the data are provided below.

• Complain: Now people are always touching each other in public. I despise that.

• Lament: It's very difficult for me to get my hopes up because there’s—I'm always waiting for the other shoe to drop. (The speaker has spent “years in Mexican prison and in legal limbo” and has endured “[l]ost or contaminated evidence, missing witnesses and agonizing delays”.)

• Evaluate Positively: [Mary] was always trying to encourage people to live a better life. (The speaker “had a particularly strong bond with Mary,” her “mentor,” who “inspired” her.)

• Describe: [A]s a writer, you have to recognize that nothing lasts and things change, that there’s no one time in the history of publishing where everything was one way, and then all of a sudden there was change. It’s always changing. (The speaker is arguing that change is normal and writers should adapt rather than get upset.)

Tokens were coded as Complain if they conveyed irritation and/or anger regarding the action. Typically, the grammatical subject was the target of the negative emotion, but this was not essential. Expressions of negative emotion also include laments. Tokens were coded as Lament if the emotion expressed is sadness or regret, not anger, and the goal is to elicit or express sympathy, not to complain or assign blame. (The situation could be one in which no agent is responsible, the agent is unknown or not focused on, or responsibility is diffused across many agents.) The subject was the target of the sympathy nearly 70 percent of the time.

Always progressives are also used to evaluate people/things positively. Depending on the circumstances, this could be equivalent to expressing approval, appreciation, admiration, and/or (with agentive referents) praise. Tokens that serve these functions were labeled as Evaluate Positively (henceforth simply “Positive”). See (3) for an example. All but four of the 56 Positive tokens involved praising humans and in all but six the grammatical subject is what was evaluated.¹⁰ The final category is Describe, referring to emotionally neutral statements such as (4). This label was applied if no indications of negative or positive affect were present. In such cases, the progressive had aspectual meaning (describing action in progress, ongoing states of affairs, etc.).

In all cases, to determine a token’s function, I looked at the token, the KWIC view, and the expanded context. That is, this was the default procedure and carried out even if the category seemed clear from the token or KWIC view alone. In some cases it was also necessary to consult the source text. Context is crucial because tokens considered in isolation can be ambiguous or misleading. Consider the token Katherine was always smiling. This was coded as a complaint because shortly before that, readers are told that Rosemary, from whose perspective the story unfolds, is “determined not to let Katherine get under her skin” like Katherine usually does. Other parts of the novel reveal that Rosemary is quite jealous of Katherine, whose perfection “ma[kes] Rosemary feel inferior.”

Like the coding of semantic prosody, the coding of discourse functions is not something that can be automated; it requires a human coder who interprets the data carefully and holistically rather than relying on simple cues such as which verb appears in the token. Such a process necessarily involves a degree of subjectivity. However, the categories and coding method described above have been honed across multiple studies.

Coding of Subject Type

The tokens were further coded according to their grammatical subjects. The categories are defined below.

• First person pronoun (1PP): I and we used to refer to humans. In two cases, additional words followed (we the patients and we coaches).

• Second person pronoun, addressee (Addr): You (singular or plural, including non-standard plurals) used to refer to a human addressee or addressees.

• Second person pronoun, generic (Gen): You (singular or plural, including non-standard plurals) used to refer to human “generic you,” as in (1–3) below.

• Third party (3P): The pronouns he, she, and they and other nominals used to refer to third party humans. This includes proper names, kinship terms, noun phrases, and relative pronouns such as who.

• Nonhuman (NH): Nominals, including personal pronouns, used to refer to nonhuman subjects (e.g., they used to refer to those stories, pens, and my presentation).

As seen above, 2PP subjects were split into two categories: Addressee you, used to directly address people, and generic you, which refers to people in general, as in (1), or to a subgroup that the speaker is a member of, as in (2-3).

• 1. Everybody does mistakes throughout your career, because you're always trying to do different things. (Describe)

• 2. As a coach, you're always trying to figure it out. (Describe)

• 3. In the service […] you're always living out of bags. It’s so aggravating. (Complain)

The test used to distinguish humanness from nonhumanness for subjects other than first or second person pronouns or the relative pronoun who was if one could answer “yes” to “is X a human?” or “are Xs humans?” For example, left wing voters and journalists pass this test while the Left and the media do not. Exceptions were made for things with human-like levels of sentience (examples from the data include God, the devil, and aliens engaged in an arms race with humans), but such exceptions were rare. Relative pronouns other than who were considered human if their antecedents passed the test above. This happened three times, such as when the antecedent of that was helicopter moms. Tokens (eight) whose subjects consisted of conjoined elements from mismatching categories were discarded as uncategorizable.

Data Analysis

The distributions of functions and of subject types were tested using the chi-square goodness-of-fit test, and the relationship between function and subject type was tested through a configural frequency analysis (CFA). Explained in Von Eye (2001) and Gries (2009: 240–252), this technique computes the statistical significance of the global chi-square as well as the contributions of each configuration—in this case, each function-subject combination—and identifies those that appear significantly more often than expected (called “types”) or less often than expected (“anti-types”). This was calculated using the R script HCFA v. 3.2 (Gries, 2004) with Holm’s correction for multiple post hoc (binomial) tests (i.e., the sequentially rejective Holm-Bonferroni test).

Results

The method yielded 2,535 hits, of which 1,233 were retained as tokens. The results of the statistical analyses are described below.

Goodness-of-fit Tests

The distribution of tokens across functions and subject types is given in Table 1 as raw frequencies and percentages of the total. Most (74.0%) of the always progressives were used in neutral descriptions. Another 16.4% served as complaints, and only 5.1% and 4.5% served as laments and positive evaluations. The most common subject type was 3P (53.0% of tokens). About half as many subjects (24.5%) were 1PP, 14.2% were NH, and 8.4% were 2PP.

Table 1 Distribution of tokens across categories

Chi-square goodness-of-fit tests were performed comparing the distribution of functions and subject types to hypothesized equal distributions. Deviation from an equal distribution was highly significant (P < 0.001) in both cases: for function χ2 (df = 3) = 1620.70 and for subject type χ2 (df = 4) = 1012.04.

Configural Frequency Analysis

The distribution of functions among tokens of each subject type were, overall, more similar than different (Fig. 1) and resembled the distribution of functions in the data set as a whole.

Fig. 1

Functions of always progressives by subject type

Nevertheless, each subject type was linked to a unique behavioral profile, as revealed by the CFA. This procedure includes running a chi-square test of independence, which showed that the relationship between function and subject type was significant: χ2 (9, N = 1233) = 121.16, P < 0.001. The CFA also determines which configurations (e.g., 1PP-Lament) contribute the most to the global chi-square. That is, rather than only comparing the overall distribution of subject types and functions to hypothesized equal distributions, the CFA also conducted a similar analysis for each of the 20 possible combinations (configurations) of subject types and functions. As mentioned earlier, subject-function configurations (e.g., 3P-Complain) that appear more or less often than we would expect to see based on chance, alone, are called types and anti-types, respectively.

The full CFA output (see Table 2) includes the following information regarding each configuration: observed and expected frequencies; the chi-square contribution; a greater- or less-than sign indicating that a configuration exceeds or falls below its expected frequency (is a type or anti-type); a probability value calculated using the Holm-Bonferroni adjustment; a significance decision at the P = 0.05, 0.01, or 0.001 level (indicated by one, two, or three asterisks); and the co-efficient of pronouncedness Q (a measure of effect size).

Table 2 CFA output

The CFA revealed three types and five anti-types. The types were 3P-Complain (P < 0.001), 3P-Positive (P < 0.05), and Addr-Complain (P < 0.05). The anti-types were 1PP-Complain (P < 0.001), 1PP-Positive (P < 0.001), NH-Complain (P < 0.01), 3P-Describe (P < 0.05), and Gen-Complain (P < 0.05). For convenience, this subset of information from Table 2 is presented separately, in Table 3.

Table 3 Types and anti-types

Discussion

In this section, I propose explanations for the function-subject relationships shared above. Because the data consisted of present day American English, it is possible that the discovered usage patterns are not representative of what would be found in other dialects or languages and that, even if they are, that the explanations I offer for those patterns are not valid for speakers of other dialects or languages, who may adhere to different social norms. Whether speakers with similar expectations regarding discourse functions such as praise and criticism use progressives or their equivalents in a manner to that of speakers of American English is yet to be determined.

The overall distribution of functions of always progressives in this study closely aligns with that in Lindley (2020), and the distribution of subject types roughly aligns with Römer’s (2005: 69) findings regarding progressives in spoken British English. Speakers use always progressives flexibly for a variety of purposes because—as argued in Lindley (2020)—the features for which they are named make them useful syntactic vehicles for complaints but also enable speakers to praise others effusively, lament their misfortunes, or describe in an emotionally neutral manner things that are continually true. The results show that which of these functions a token serves is at least partly related to its subject. The remainder of this section is divided into two main parts: I first discuss subject-function relationships involving complaints and then discuss those involving positive evaluations. I discuss those involving positive evaluations.¹¹ Broadly speaking, I argue that the observed types and anti-types are the product of multiple interacting social and practical concerns naturally associated with who/what a speaker is talking to/about.

Complaints

This section discusses the types (Addr- and 3P-Complain) and anti-types (Gen-, 1PP-, and NH-Complain) involving complaints. In 189 of the 202 complaint tokens (93.6%), the grammatical subject was also the thing evaluated. These results suggest an inclination to use always progressives to complain about other people and a disinclination to complain about ourselves or nonhumans. I argue below that, more specifically, the usage patterns regarding complaints can be attributed largely to the usefulness of complaining to/about others and unusefulness of complaining about ourselves and non- or less agentive things. Additionally, regarding 2PP subjects, I argue that the contrasting ways that speakers use addressee and generic you demonstrate the methodological necessity of considering them separately.

Type Addr-Complain

The preponderance of direct criticism of addressees is surprising given its impoliteness, but it seems that its potential high efficacy can override such concerns. In politeness theory, negative face is “freedom of action and freedom from imposition” and positive face is one’s positive self-image, tied to the desire to be liked by others (Brown & Levinson, 1987: 61). Complaining directly to someone about their habits threatens both types of face because it is an imposition (it implies that the addressee should alter their behavior) and is critical. This effect is exacerbated in always progressive complaints because words like always are associated with exaggeration (Claridge, 2011: 51)—which, in complaints, “emphasize[s] the problem or the failings of the addressee” (Legitt & Gibbs, 2000: 7). Rude complaints involving exaggeration elicit even more anger, disgust, and scorn than those involving sarcasm (ibid.: 8, Table 1), which is “almost exclusively negative and hostile” (ibid.: 10).

Despite being rude—in fact, precisely because they are—complaints directed at addressees can be highly effective. First, in lodging the complaint directly with the offender, the complainer ensures that that person becomes aware of the problem. A positive outcome is not guaranteed, but it becomes possible. Second, the unpleasantness and heightened “emotional impact” of complaints involving exaggeration can make them “very memorable” and thus can lead to their having “long-ranging effect[s]” (Claridge, 2011: 142). Third, the complainer’s willingness to do something that is borderline socially unacceptable may convey that a situation is serious enough to merit such drastic action. In sum, complaining to someone’s face can work quite well, but one obtains the benefits at a cost: The social relationship may be irreparably damaged.

Type 3P-Complain

Complaining to but not about an addressee occurs during activities such as gossip, venting, and commiseration. 3P-Complain is a type because of the importance of such activities to humans, who—compared to other species—rely heavily on social information for their survival and well-being. The value of gossip, in particular, is well documented. Gossip is “the exchange of information about absent third parties” (Foster, 2004: 81) combined with sharing a positive or negative evaluative stance toward that information (82), with the most common understanding being that this stance is negative. Gossip is pervasive, accounting for between about 66% (Dunbar, 2004: 100) to 80 or 90% of daily talk (Emler, 1994: 131). Our species even has a cognitive bias that promotes gossip: Baumeister et al. (2001: 360) describe a body of research showing the existence of a negativity bias, active in various domains, whereby we give more weight to unpleasant phenomena. This includes noticing negative things more, remembering them better, and experiencing them as more “potent.” In addition, we are more likely to socially transmit information if it is negative (Bebbington et al., 2017).

We have evolved to gossip because it serves important social functions benefitting groups and individuals (Beersma & Van Kleef, 2012; Dunbar, 2004; Kniffin & Wilson, 2010; etc.). These functions include (1) policing people while protecting people’s face, (2) sharing or validating crucial social information, and (3) bonding. Complaining about third parties can also (4) boost our health and (5) be a powerful weapon against enemies. I discuss these five functions below.

First, despite their passive-aggressive nature, complaints about third parties can successfully stop irritating actions. Negative gossip often makes its way back to and has repercussions for the guilty party, and the ensuing shame or damage to reputation can lead them to change their behavior. This is the “policing function” of gossip (Dunbar, 2004: 103), also called the “group protection” function (Beersma & Van Kleef, 2012), whereby gossip keeps people in line with social norms and, importantly, protects social groups against “free-riders,” who “take the benefits of sociality but decline to pay all of the costs” (ibid.:106).

Importantly, gossip serves this protective function without requiring the complainer to be confrontational. As stated earlier, complaining about someone to their face is direct but rude. Our species is “adapted for group living” to the point of “obligatory interdependence” (Brewer & Caporael, 2006: 143), so we feel strong pressure to behave in ways that protect the interpersonal relationships that sustain us. This does not eradicate impoliteness, but in certain situations, at least, we avoid face-threatening acts. According to Wolfson’s (1988) bulge theory, we more easily get away with being impolite to strangers and intimates, on the far ends of the scale of closeness, than we do to people in the middle (cf. Dunbar, 2004: 107). Even people who are blunt with telemarketers probably think twice before criticizing their boss or a colleague to their face. In comparison, gossiping about those people in their absence is far less risky socially and could still be effective in stopping unwanted behavior.

One scientific view of gossip is that it is a way to trade and/or confirm potentially valuable information, its second crucial social function. Explaining their “theory of the human self as information agent,” Baumeister, Maranges, and Vohs begin from the premise that “much of [our] social and solitary activity is devoted to dealing with information”—including “seeking and acquiring [it],” “communicating one’s thoughts to others,” and “circulating information through the group” (2018: 36). This often includes information about group members, such as who is dishonest, who is seeking a partner, etc. Possessing this knowledge helps us survive and thrive socially, whereas those who are “out of the loop” are severely disadvantaged.

A third function of gossip is that it facilitates bonding. Dunbar theorizes that language in general and gossip in particular originated as a bonding mechanism that was more efficient than physical grooming and thus could work for increasingly large primate groups (2004: 100). Whereas grooming is a time-consuming, one-on-one activity, gossip enables multiple individuals to bond at once (102). In modern times, for many people, gossip is a pastime that helps us “have a good time with others” (Beersma & Van Kleef, 2012: 2645) and that thereby spawns and sustains friendships. A study by Grosser et al. (2010) showed that negative gossip occurred more between co-workers if they were also friends.

Complaining about third parties serves a fourth function: In the context of venting and commiseration it can boost our physical and mental health. Venting about one’s problems is a way to elicit validation (Norlock, 2018: 129), “confirmation that one is not alone” (133), and empathy. Ventilation, validation, and empathy “directly reduce arousal” in a distressed person and may even boost self-worth (Thoits, 2011: 154). In contrast, suppressing one’s emotions is linked to numerous negative physiological and psychological side effects (Kowalski, 2002: 1029). Even or perhaps especially if we lack control over something, swapping complaints about it makes it more bearable (Norlock, 2018: 119), partly because it facilitates “sympathetic bonding” (129). By strengthening social bonds, which “are positively and causally related to mental health, physical health, and longevity” (Thoits, 2011: 154), complaining can significantly contribute to our well-being.

Fifth and finally, some gossip is purely malicious and wielded like a weapon. If one’s goal is not to stop a behavior but to harm someone’s reputation, direct confrontation is useless—especially if the target is innocent. Malicious gossipers are assisted in their endeavor by the aforementioned negativity bias. In sum, complaining about third parties has served important social functions since the earliest days of language, to the point of humans evolving a cognitive bias toward spreading gossip, and this explains its continued pervasiveness in our lives—including its being a common function of always progressives containing 3P subjects.

Anti-type Gen-Complain

Generic you subjects slightly outnumbered addressee you subjects (58 versus 45), and the two subject types behaved differently: Most notably, complaints were far less common with generic you subjects (1.7%) than with addressee subjects (37.8%). In addition, all 17 Addr-Complain tokens consisted of negative evaluations of the grammatical subject, whereas the single Gen-Complain token was not about the subject.

The anti-type status of Gen-Complain stems from the semantics of generic you. Whether referring to people in general or a subgroup containing the speaker, generic you is more akin to a 1PP plural than to addressee 2PP. Speakers using generic you could just as easily use we, and in fact speakers sometimes alternate between the two options, as in (4).

• 4. In the service, we're constantly going from place to place. And you're always living out of bags.

The speaker uses a 1PP plural subject in a constantly progressive, then switches to generic you in an always progressive, with no change in the referent.

Generic you is a second person pronoun in form only, not semantically or functionally. This is why the distribution of functions associated with generic you resembles that associated with 1PP subjects (see Figure 1) and why, like Gen-Complain, 1PP-Complain is an anti-type. The dramatically contrasting behavioral profiles of generic and addressee you show that—certainly in any study in which empathy (point of view) may be important—one must analyze the two types separately.

Anti-type 1PP-Complain

Complaining about ourselves is of little use and even harmful, as sharing negative information about ourselves can damage our reputations.¹² The anti-type 1PP-Complain indicates that we are averse to doing this. A qualitative analysis of the tokens indicates this even more strongly—yet simultaneously shows that more research is needed before we can fully understand this anti-type.

Complaining about oneself is even rarer than the low number of 1PP complaints indicates. 1PP complaints were exceptional among the complaint tokens in that only half (ten)—as opposed to 93.6% of the Complain tokens as a whole—were complaints about their grammatical subjects. In (5), for example, a speaker uses a 1PP token to complain about not himself but his mother.

• 5. Why don't you sometimes, once in a while, close the doors of the cupboard?—I'm always closing them (slam, slam). […] God, Mom, you can be so annoying.

Moreover, of the ten 1PP complaints that were about their subjects, five contained plural subjects, referring to society or a group rather than the speaker only. The remaining five contained and were about 1PP singular subjects, two of which involved speakers criticizing themselves, as in (6). In the other three, the speakers are relaying complaints made by other people, as in (7).

• 6. I always thought [life] was a science. […] I was always looking for the math. As a parent, as a husband, as an actor, just as a human being, I suck a lot. You know, I suck so much more than I’ve thought that I should at forty-six. (Hollywood actor Will Smith)

• 7. [My teachers] say I am always staring into space with my mouth open, but really what's happening is that I'm thinking of some question I want to ask.

In sum, the data contained 20 1PP-Complain tokens, but a mere two instances of people complaining about themselves. This is strong evidence that we avoid complaining about ourselves, yet does not explain why 1PP complaints like (5) were not more numerous.

Anti-type NH-Complain

The anti-type NH-Complain can be attributed to the nature of NH subjects: Low in agentivity, they are unlikely to be subjects of progressive verbs (which prefer subjects that are agentive, human, volitional, animate, etc.), and when they nevertheless are, the actions attributed to them are unlikely to be the sort that merits anger. Much more often than appearing in complaints, NH subjects appeared in Describe tokens consisting of bland statements about innocuous actions, for example, “something is always happening” (a statement nearly devoid of content), “time is always moving forward,” and “[subject] is always changing.” NH subjects said to be “always changing” included the landscape, nature, culture, the world, and the universe. These general, timeless truths are unlikely to anger people, and even if a person did wish to assign blame, non-agentive things like “time” and “landscapes” are not logical targets.

Positive Evaluations

The CFA identified one anti-type and one type involving positive evaluations. The anti-type was 1PP-Positive (P < 0.001, Q = 0.01). The data contained a single 1PP-Positive token and, in it, the speaker is sharing credit with others. (The subject, we, refers to an administrative team.) The type was 3P-Positive (P < 0.05, Q = 0.014). I argue below that both findings reflect the desire to avoid bragging and appear likable. In addition, I describe two subfunctions of 3P-Positive tokens.

Anti-type 1P-Positive

1PP-Positive tokens were nearly non-existent because, even in individualistic societies, bragging about oneself is frowned upon. People view self-promoters as braggarts and less likable (Scopelitti, Loewenstein, & Vosgerau 2015). Women in particular feel compelled to conform to bragging-related modesty norms due to fear of backlash (Lindeman et al., 2019). Bragging can also be beneficial; studies by Chaudhry and Loewenstein (2019) and Heck and Krueger (2016) show that it can lead to increased perceived competence. Yet those same studies show that this benefit is limited: In the first study, the increase came with a negative effect, a decrease in perceived warmth (19). In the second, brags about competence boosted perceived competence only if no evidence contradicted them; otherwise, perceived competence was reduced. Additionally, brags about being moral (even if accurate) led to decreased perceived morality. Finally, in the study of Scopelitti, Loewenstein, and Vosgerau (2015: 911), bragging did not boost judgments of successfulness. Since bragging often brings negative consequences, it is natural that speakers avoid bragging using always progressives.

Type 3P-Positive

The emotionality that can be conveyed via always progressives is not restricted to negative emotions; the construction can also be used for effusive positive evaluations. This function was so common among tokens with 3P subjects that 3P-Positive was a type. Acknowledging the good qualities of others can be important and necessary, but 3P-Positive tokens may be over-represented in the corpus (compared to in English in general) due to so many source texts containing interviews conducive to evaluations of others. Nearly three quarters (35 out of 47) of the tokens coded as 3P-Positive were direct or indirect quotes of people being interviewed for news articles, etc. Interestingly, over half (19 out of 35) of these interview tokens served one of two subfunctions.

First, in news stories about deaths, interviewees used the tokens to paint a flattering picture of the deceased (“eulogy” function). Second, in stories about the interviewee, interviewees used the tokens to acknowledge others who had contributed to their success (“credit/thanks” function). Table 4 displays the number of 3P-Positive tokens appearing in interview contexts along with their subfunctions and the number appearing in non-interview contexts.

Table 4 Types of 3P-Positive tokens

In many cases, the explicit purpose of a story was to celebrate someone’s achievements, as in the story “The greatness of Larry Fitzgerald, as seen through the eyes of his [quarterbacks].” It is unsurprising, then, that most (62.9% or 22 of 35) of the 3P-Positive interviewee tokens were evaluations of the topic (always a person) of the interview. What is surprising is how often the interview topic was a deceased person—often a murder victim. In two of the nine tokens serving the eulogy subfunction, the grammatical subjects were famous people who died of natural causes. The other seven eulogy-function tokens involved violent deaths: four murders, the suicide of a prominent actor, a murder-suicide, and a hit-and-run.

Although the eulogy function can appear in non-interview contexts (such as actual eulogies),¹³ the deaths of famous people and murder victims are especially likely to be covered by the media and may also be especially likely to elicit praise. Deceased famous people are praised because they tend to have achieved many things (and have wide social circles), and murder victims are praised because emphasizing the good qualities of victims helps loved ones convey and validate the magnitude of their emotions in the wake of a traumatic experience: The more generous, moral, talented (etc.) the victim, the bigger the loss and injustice, and the more valid one’s grief and anger. Making such comments publicly could also serve as a form of (weak) revenge in that sanctifying the victim implicitly vilifies the murderer in equal measure. It could even incite the public to act in some way (e.g., to track down a suspect or to support relevant new legislation).

In contrast to the eulogy tokens, none of the ten credit/thanks tokens were about the interview topic. Instead, interviewees used these tokens to publicly acknowledge another person’s role in their (the interviewee’s) success. In one story, a rising college football star told reporters that “he owes much of his success this season to [his teammate],” who is “always teaching [him]” valuable skills. In another, a student asserts that family support is crucial for success in higher education right before sharing that “[her] mom was always sending [her] great messages that encouraged [her]” and was her “personal cheerleader.”

The credit/thanks tokens all occurred in stories about the interviewees and, in most cases, their achievements—a context in which the risk of seeming arrogant is high. Arrogance can damage social bonds, whereas perceived humility is linked to both forging and repairing them (Davis et al., 2013). Humbly conferring credit on others, the opposite of bragging, makes people seem warmer—albeit less competent (Chaudhry & Loewenstein, 2019: 19). Although both aspects of one’s reputation must be maintained, someone whose achievements are currently being publicly celebrated need not worry about seeming competent and would be wise to focus on being likable.

The public nature of the credit/thanks and eulogy tokens enhances their efficacy. Consider award speeches: A person who contributed immensely to an Academy Award winner’s success and had already been earnestly thanked in private might be offended if they were not also acknowledged in the televised acceptance speech, which makes the thanked person look good to an enormous number of people. Likewise, grief can incite community action only if the community knows about it.

Concluding Remarks

Nominals encode basic information about the discourse context, that is, who/what (and what kind of person/thing) is being spoken to and/or about. Pronouns are especially rich in this regard because they are marked as first, second, or third person. This study shows that grammatical person and humanness are related systematically to the functions of always progressives. The observed subject-function relationships arise as natural consequences of humans being agentive and of practical considerations related to social/interpersonal goals such as protecting positive/negative face, appearing likable and humble yet competent, and bonding. In interviews, two useful subfunctions of the Evaluate Positively function were common: a credit/thanks function and a eulogy function. In these cases, speakers take advantage of always progressives’ potential to be emotional and hyperbolic as a way to heap praise on those who have helped them or on lost loved ones, thereby effusively expressing gratitude or grief. Naturally, one must tread carefully when interpreting correlations, but one strength of the analysis is that it draws heavily on evidence from fields outside of linguistics such as psychology and sociology. Future research in these and other fields on complaints, exaggeration, acknowledgements, and bonding should enable us to refine and expand what is offered here.

Features of nominals such as grammatical person and humanness merit more frequent inclusion in studies of constructions’ discourse functions. The findings reported here that are, in hindsight, unsurprising (e.g., the scarcity of positive evaluations of 1PP subjects) demonstrate this point the most strongly in that these intuitive connections between functions and nominals are nevertheless rarely empirically tested—if they are considered at all. Genre is a go-to variable in corpus studies of linguistic variation, yet we commonly neglect to consider the influence of a factor as important as whether we are talking about, say, ourselves versus a rock. At the very least, when studying a construction, we should consider examples containing a range of types of nominals. Finally, one rather specific takeaway is that we must exercise caution regarding the second person pronoun. Generic you, which is very common, is semantically and functionally more like a 1PP. Functional analyses that conflate generic and addressee you likely obscure useful data patterns.